Mimic Robotics

div]:mx-auto">Today we are proud to announce the mimic hand M1, our highly backdrivable, tendon-driven robotic hand designed and built in-house for industrial automation. Alongside it, we are introducing the mimic wearable U1 (“umimic”), our exoskeleton that captures human demonstrations directly matched to the kinematics of mimic hand M1. Together with our fully in-house developed software stack, these systems provide an integrated platform to enable general-purpose dexterous manipulation.<br>From day one, mimic has been laser-focused on the single goal of general-purpose dexterous manipulation. We believe that the only way to solve general-purpose manipulation while creating value at scale is to take the “full-stack approach”. Deploying autonomous fleets of robots will depend on vertically integrated stacks with full hardware observability, just as full vertical integration mattered in autonomous driving.<br>Every serious attempt to bring foundation models to the physical world runs into the same wall: robotics has no equivalent to internet-scale training data. So an optimal approach must draw on multiple sources and combine data across qualities and quantities. What pre-training delivers is representations - visual, semantic, behavioral, and physical. The largest source of behavioral and physical priors at meaningful scale is large-scale video of humans. For post-training, nothing beats the robot: live deployment data and teleoperation, collected on the target hardware in the target environment.<br>However, the embodiments of traditional two-finger gripper robots and the human embodiment are fundamentally mismatched. If we ground pre-training in human video and then deploy a two-finger gripper, we introduce a cross-embodiment gap that renders our pre-training and post-training phases misaligned: the end effector is different, and the discrete steps a human takes to manipulate an object are not the same a two-finger gripper would need to take.<br>We never introduce this cross-embodiment gap in the first place. We hold morphology constant across every phase of learning by running the entire pipeline on hands. This is the bet we are making at mimic.<br>To balance our human data needs across qualities and quantities, we developed a strategy based on a “data pyramid”. At the base of the pyramid sits human video data: lower quality per sample, but available at massive scale. In the middle, wearable device data: higher quality than pure arbitrary egocentric video, but easier to scale than robot teleoperation. And only at the top, robot teleoperation and deployment data: the highest quality of data, collected directly with our own robots.

mimic hand M1<br>Solving general-purpose dexterous manipulation requires a general-purpose robotic hand, one that covers the full range of human capability: precise and sensitive tasks, heavy payloads, and resistance to severe collisions, all with the same system. The mimic hand M1 is our answer, designed and manufactured in-house. We built it around an AI-first architecture, prioritizing what actually matters for real-world robot learning and industrial use: a 1:1 human form factor, highly backdrivable joints, bi-directional actuation, robustness, reliability, and sustained strength.

Watch Video

AI-first architecture<br>AI-first means designing for and with the data. We matched the robot’s joints to the most functional degrees of freedom of the human hand, including abduction and an opposable thumb, so that human demonstration data transfers cleanly onto the robot. The M1 has 15 active degrees of freedom across 21 joints. That number wasn't arbitrary: it reflects insights from data we've collected on industrial tasks. Where a joint contributed little to real task performance, we dropped it in favor of higher long-term reliability and a simpler mechanical design.<br>Physical AI changed the design requirements of robotic hardware. In contrast to conventional automation, learned manipulation requires adaptive, compliant, and inherently force-sensing systems that provide AI models with rich contact information that vision alone cannot provide. Quadrupeds and early humanoids showed that an effective way to realize these system characteristics is through the use of highly backdrivable actuators, where every actuator doubles as a fast, precise, and safe force sensor.<br>We transferred that approach to the M1, realizing highly backdrivable joints with a backdriving torque under 0.05Nm, low enough to sense weights as light as 50g directly through motor current. Full joint encoders, bi-directional actuation, and low backlash ensure that the M1 has full observability and precise control over every move it makes. A high-speed dual-encoder setup, on the motor and on the joint, lets us resolve uni-directional contact forces down to 0.1N. On top of that, dedicated fingertip tactile sensors add tangential shear-force sensing and higher spatial resolution where it matters most. The M1 is the sensor...